AI Governance Tracker preview

Methodology

This page explains what the tracker covers, where its records come from, how they are checked, and how to read the composite index. Data as of the latest check.

What we track

The tracker follows 10 jurisdictions: Brazil, Canada, China, the European Union, India, Japan, Singapore, South Korea, the United Kingdom and the United States. The European Union is treated as one jurisdiction because its AI rules apply across all 27 member states; national rules inside member states are not yet covered. US coverage includes federal instruments and selected state laws that are widely cited in AI policy debate.

There are currently 53 instrument records. An instrument is included when it is mainly about artificial intelligence, or when it contains provisions written specifically for AI or automated decision-making. Each record is one of these types:

Sources

Every record links to a primary source wherever one exists: an official gazette, legislature, ministry, regulator or government press office. Secondary reporting is used only to find instruments, never as the sole basis for a record's status or date. Each record shows the date it was last checked against its source.

Verification

Each record carries a verification label shown on the map:

Current counts: 46 verified, 6 partially verified and 1 unverified.

Status

Laws change over time, so every record has a status: in force; enacted or adopted but not yet applicable; proposed; revoked; withdrawn; lapsed; or not introduced. Revoked, withdrawn, lapsed and not-introduced records stay in the list as history but do not count toward the index. A rule that is in force counts for more than one that has been passed but does not apply yet, and both count for more than a proposal.

How to read the composite index

Each jurisdiction gets a preliminary composite score from 0 to 100. A higher score means a more demanding environment for organisations that build or deploy AI there: more binding rules that already apply, clearer enforcement, and more legal change still in motion. Government support measures, such as national strategies and funding programmes, are taken into account too. A lower score means fewer binding obligations so far. The score is not a ranking of how good or bad a jurisdiction's policy is.

The score is calculated only from the public instrument records listed on the map, using a fixed, documented internal model that is applied the same way to every jurisdiction. It has diminishing returns, so a long list of instruments cannot by itself push a score to the top of the scale. We do not publish the model's internal formula; the records behind every score are public, so anyone can review the evidence.

On the map, scores are grouped into bands for readability:

Small differences between scores (a point or two) are within the uncertainty of the method and should not be read as meaningful.

Office-holders

A later version will add the public officials who hold AI-relevant offices in each jurisdiction, such as ministers and regulator heads, drawn from open public-domain data with start and end dates. Until that data is in place, every jurisdiction is treated the same on this point and it does not separate the scores. The tracker is a policy reference, not a screening or due-diligence tool.

Limits

Updates and corrections

Records are rechecked against their sources on a rolling basis and the last-checked date is updated each time. A public route for reporting corrections will be added before launch.

Credits

Boundaries from Natural Earth (public domain), shown for reference only and not implying any position on disputed territory. Map rendering by MapLibre GL JS (BSD-3-Clause).